EXP-EVO-LOOP-FULL

… pending

EvoLoop full — NOVELTY vs RANDOM vs NO selection, 3 conditions × 3 runs

2026-04-30 L6 level paper4 9 runs $0.22

new

premortem

mock

real

metrics

analyze

review

krit

done

In plain language

Researchers are testing how different ways of guiding artificial evolution affect the variety of outcomes. They are comparing a method that encourages new and different results with methods that don't have specific goals or just randomly select. The experiment is still running, so the final findings are not yet available.

▶ Technical details

Research hypothesis

Pilot (exp-evo-loop-pilot) showed DECLINING novelty trajectory under NOVELTY selection (0.887→0.871→0.858 across gens 1-3). Full experiment isolates whether this decline is selection-specific (NOVELTY condition drives bottleneck faster) or substrate-universal (RANDOM and NO selection also decline). Falsifiable prediction: NOVELTY ≈ RANDOM ≈ NO selection trajectories → decline is substrate property, not selection artifact. If NOVELTY > RANDOM > NO: selection creates differentiation that counters archive accumulation.

Experimental setup

Type: factorial

Condition	Parameters
EVO_NOVELTY	selection_mode: novelty, n_runs: 3
EVO_RANDOM	selection_mode: random, n_runs: 3
EVO_NONE	selection_mode: none, n_runs: 3

Factors: selection (NOVELTY, RANDOM, NO_SELECTION)

Parameters

n_agents: 4
model: google/gemini-2.5-flash-lite
temperature: 0.8

evolutionary-loop novelty-selection multi-generation three-condition paper4-main